Examples

Usage examples for Maple-Syrup-Pi-Camera

The Google Coral USB Accelerator used in the Maple-Syrup-Pi-Camera can only run TFLite models that were compiled by the EdgeTPU compiler. However, there are plenty of TFLite models already available ready to be used. In addition to that, Google doesn't officially support the use of the Coral USB Accelerator with the Raspberry Pi Zero anymore, therefore I had to manually adapt the PyCoral API (already available in the SDCard image). Still, if you try a script that uses the PyCoral API, it may need small adaptations.

Using the supplied SDCard image, you can go to /home/pi/coral/ and try:

Detection:

90 different things - Using SSD MobileNet V1

python object_detection_from_cam.py --model test_data/ssd_mobilenet_v1_coco_quant_postprocess_edgetpu.tflite --labels test_data/coco_labels.txt

90 different things - Using SSD MobileNet V2

python object_detection_from_cam.py --model test_data/ssd_mobilenet_v2_coco_quant_postprocess_edgetpu.tflite --labels test_data/coco_labels.txt

90 different things - Using SSDLite MobileDet

python object_detection_from_cam.py --model test_data/ssdlite_mobiledet_coco_qat_postprocess_edgetpu.tflite --labels test_data/coco_labels.txt

Faces - Using SSD MobileNet V2

python object_detection_from_cam.py --model test_data/ssd_mobilenet_v2_face_quant_postprocess_edgetpu.tflite --labels test_data/coco_labels.txt

I

Classification:

1000 different things - Using Inception V1

python classify_from_cam.py   --model test_data/inception_v1_224_quant_edgetpu.tflite  --labels test_data/imagenet_labels.txt

1000 different things - Using Inception V2

python classify_from_cam.py   --model test_data/inception_v2_224_quant_edgetpu.tflite  --labels test_data/imagenet_labels.txt

1000 different things - Using Inception V3

python classify_from_cam.py   --model test_data/inception_v3_299_quant_edgetpu.tflite  --labels test_data/imagenet_labels.txt

1000 different things - Using MobileNet V1

python classify_from_cam.py   --model test_data/mobilenet_v1_1.0_224_quant_edgetpu.tflite  --labels test_data/imagenet_labels.txt

1000 different things - Using MobileNet V2

python classify_from_cam.py   --model test_data/mobilenet_v2_1.0_224_quant_edgetpu.tflite  --labels test_data/imagenet_labels.txt

1000 different things - Using MobileNet V3

python classify_from_cam.py   --model test_data/tf2_mobilenet_v3_edgetpu_1.0_224_ptq_edgetpu.tflite  --labels test_data/imagenet_labels.txt

Pets

python classify_from_cam.py   --model test_data/ssd_mobilenet_v1_fine_tuned_pet_edgetpu.tflite  --labels test_data/pet_labels.txt

900+ Birds

python classify_from_cam.py --model test_data/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite --labels test_data/inat_bird_labels.txt

1000+ insects

python classify_from_cam.py   --model test_data/mobilenet_v2_1.0_224_inat_insect_quant_edgetpu.tflite  --labels test_data/inat_insect_labels.txt

2000+ plants

python classify_from_cam.py   --model test_data/mobilenet_v2_1.0_224_inat_plant_quant_edgetpu.tflite  --labels test_data/inat_plant_labels.txt

Models prepared for this camera:

• MultiPose - shows how to convert a TensorFlow.js model to TFLite
• Automatic License Plate Recognition - shows how to directly modify TFLite models
• Face Mask Detection (Mask, No Mask, Poor Mask) - also useful as a base for training other object detectors

Other possible sources for TFLite models:

• Colab/Jupyter tutorials about training TensorFlow models for Edge TPU, and other tutorials
• https://coral.ai/models/all/
• https://coral.ai/examples/
• https://modelplace.ai/models
• https://github.com/Qengineering
• https://www.tensorflow.org/lite/guide/hosted_models